Charge and Discharge Control using Reinforcement Learning for Parallel Resonant PMSG System in Series Hybrid Vehicles

Abstract

A parallel resonant permanent-magnet synchronous generator (PMSG) system, which consists of a diesel engine, PMSG, full-bridge rectifier, and resonant parallel capacitors, has been proposed for series hybrid vehicles to save energy and reduce costs. In series hybrid vehicles, the parallel resonant PMSG system cannot adjust the output power; thus, charge-discharge control based solely on whether the engine is turned on or off is required. A charge-discharge control method using a state-of-charge map has been proposed that focuses on reducing the number of engine-starts. However, this conventional method is limited in its ability to extend battery life and improve fuel consumption in addition to reducing the number of engine-starts. Therefore, this study aims to further improve the charge-discharge control of parallel resonant PMSG system using reinforcement learning.